Abstract
The realization of a low-carbon economy and the large-scale transition to renewable energy encompasses some of the most pressing challenges of the era. A developing technology in the energy sector is the vanadium redox flow battery (VRFB). This is a promising long-term energy storage solution that can alleviate the intermittency of renewable sources while boasting benefits over the traditional lithium-ion centered grid structure. However, VRFBs require further enhancement before becoming dominant market players; one target area of improvement is the design of the flow fields (FF) that facilitate pressure consistency, distributed diffusion, and uniform electron transfer between the electrodes and active species ions in an electrolyte solution. Guided by their increasingly pervasive nature, I plan to analyze the effectiveness of physics-informed machine learning techniques to design and maximize these favorable characteristics in a FF, with the end goal being a method to improve VRFB performance without being cost-intensive. Specifically, I will implement a SARSA network that will iteratively improve FF design through reinforcement learning. I will train the network using a FF path generation algorithm and physical simulation of each iteration with components derived from relevant fluid dynamics. Then, I will evaluate it against the three standard FF designs. With guidance, I would love to be able to construct the optimized FF and VRFB in real life.
